The Genom of Homo sapiens.pdf

SYSTEMS BIOLOGY IN S. CEREVISIAE AND HALOBACTERIUM SP. 355tectable primary sequence homology. Rosetta employsthe following steps. (1) Proteins are subdivided into putativedomains. (2) Three- and nine-residue local structurefragments are estimated based on local sequence similarityto corresponding three- and nine-residue substratesfrom protein of known structure. (3) These precomputedlocal structure fragments are assembled into global structuresby minimizing a scoring function that favors hydrophobicburial/packaging, strand-pairing, compactness,and energetically favorable residue pairings. Sincethe predictions from Rosetta are lower in resolution thanexperimentally solved structures, they need to be supportedby other data types (microarrays, proteomics, predictednetworks, etc.). In the third, fourth, and fifth community-widecritical assessments of structure prediction(CASP3, 4, and 5), Rosetta was one of the most effectivemethods employed (Bonneau et al. 2001). We are usingRosetta to predict the three-dimensional structures of allof the Halobacterium sp. proteins and protein domains ofunknown function that are less than 150 residues inlength.When sequence similarities by PSI-BLAST or Pfamhmmersearches have statistically marginal relationships,structure-based methods such as Rosetta can help confirmor negate the putative relationships. For example, aPfam search on the Halobacterium protein VNG0511Hyields a weak hit (homology) (e-value: 0.61) to PF00392,a family of bacterial transcription factors. The Rosettapredictedstructure for this protein, which has a strongmatch to a CATH family of helix-turn-helix DNA-bindingproteins (CATH I.D.: 1.10.10.10), supported this initialtentative predicted function. CATH, like SCOP, is ahierarchical classification of protein domain structures.Thus, the agreement of these two procedures (weak Pfamand Rosetta) imparts a higher degree of confidence to thisprediction not obtainable with either individual method.We can use the knowledge that VNG0551H is a DNAbindingprotein to tentatively assign the role of a putativemediator of the changes observed in the gene(s) downstreamto its position in the inferred gene regulatory network.Likewise, all function annotations that point to regulatoryor DNA-binding roles should be examined in thecontext of one or more gene regulatory networks.A second example, the protein VNG1302H, has a weakBLAST match (e-value: 0.091) to one of three domains ofa protein disulfide isomerase from Cricetulus griseus.The Rosetta-predicted structure for the proteinVNG1302H had a strong match to 1A8L (the structure fora protein disulfide oxidoreductase from the hyperthermophilePyrococcus furiosus), which has a thioredoxin-A. B.trxB2VNG0450C trxB1_1VNG5220GtxrB3VNG1302HModel 1:1A8L-02(Oxidoreductase):msrAtrh_1trh_2VNG5077GtrxB1_2argStrh1VNG2468CVNG2115HVNG1012HVNG2310H VNG1546HtrxA1_1trxA2trxA1_2VNG1302HVNG5221GVNG0711CVNG5076GFigure 6. Structure/function annotation for VNG1302H. (A) The physical interaction network for the Halobacterium sp. VNG1302H.Blue edges indicate domain fusion type interactions, green edges are yeast interaction data mapped onto Halobacterium sp. via theCOG database. VNG1302H is indicated with the sign of the rising sun. All nodes shown have one or more direct connections toVNG1302H. Proteins/nodes in this subnetwork that are known to have redox-related functions are indicated in bold. These functionsinclude thioredoxin, thioredoxin-like, and a peptide methionine sulfoxide reductase (msrA). (B) Rosetta structure prediction forVNG1302H and the closest match in the PDB to this model. This fold similarity (note the structural alignment of two key cysteines,indicated by black spheres) indicates a thioredoxin (redox) function.